allow for overriding use of EMA during sampling in decoder trainer with use_non_ema keyword, also fix some issues with automatic normalization of images and low res conditioning image if latent diffusion is in play

* migrate to conditioned prior

* unify reader logic with a wrapper (#1)

* separate out reader logic

* support both training methods

* Update train prior to use embedding wrapper (#3)

* Support Both Methods

* bug fixes

* small bug fixes

* embedding only wrapper bug

* use smaller val perc

* final bug fix for embedding-only

Co-authored-by: nousr <>

dalle2_pytorch/__init__.py

@@ -1,6 +1,6 @@
 from dalle2_pytorch.dalle2_pytorch import DALLE2, DiffusionPriorNetwork, DiffusionPrior, Unet, Decoder
 from dalle2_pytorch.dalle2_pytorch import OpenAIClipAdapter
-from dalle2_pytorch.train import DecoderTrainer, DiffusionPriorTrainer
+from dalle2_pytorch.trainer import DecoderTrainer, DiffusionPriorTrainer
 
 from dalle2_pytorch.vqgan_vae import VQGanVAE
 from x_clip import CLIP

dalle2_pytorch/dalle2_pytorch.py

@@ -61,6 +61,9 @@ def default(val, d):
 def cast_tuple(val, length = 1):
     return val if isinstance(val, tuple) else ((val,) * length)
 
+def module_device(module):
+    return next(module.parameters()).device
+
 @contextmanager
 def null_context(*args, **kwargs):
     yield
@@ -901,6 +904,7 @@ class DiffusionPrior(BaseGaussianDiffusion):
         self.channels = default(image_channels, lambda: clip.image_channels)
 
         self.cond_drop_prob = cond_drop_prob
+        self.can_classifier_guidance = cond_drop_prob > 0.
         self.condition_on_text_encodings = condition_on_text_encodings
 
         # in paper, they do not predict the noise, but predict x0 directly for image embedding, claiming empirically better results. I'll just offer both.
@@ -914,8 +918,10 @@ class DiffusionPrior(BaseGaussianDiffusion):
         self.training_clamp_l2norm = training_clamp_l2norm
         self.init_image_embed_l2norm = init_image_embed_l2norm
 
-    def p_mean_variance(self, x, t, text_cond, clip_denoised: bool):
+    def p_mean_variance(self, x, t, text_cond, clip_denoised = False, cond_scale = 1.):
-        pred = self.net(x, t, **text_cond)
+        assert not (cond_scale != 1. and not self.can_classifier_guidance), 'the model was not trained with conditional dropout, and thus one cannot use classifier free guidance (cond_scale anything other than 1)'
+
+        pred = self.net.forward_with_cond_scale(x, t, cond_scale = cond_scale, **text_cond)
 
         if self.predict_x_start:
             x_recon = pred
@@ -933,17 +939,17 @@ class DiffusionPrior(BaseGaussianDiffusion):
         model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
         return model_mean, posterior_variance, posterior_log_variance
 
-    @torch.inference_mode()
+    @torch.no_grad()
-    def p_sample(self, x, t, text_cond = None, clip_denoised = True, repeat_noise = False):
+    def p_sample(self, x, t, text_cond = None, clip_denoised = True, repeat_noise = False, cond_scale = 1.):
         b, *_, device = *x.shape, x.device
-        model_mean, _, model_log_variance = self.p_mean_variance(x = x, t = t, text_cond = text_cond, clip_denoised = clip_denoised)
+        model_mean, _, model_log_variance = self.p_mean_variance(x = x, t = t, text_cond = text_cond, clip_denoised = clip_denoised, cond_scale = cond_scale)
         noise = noise_like(x.shape, device, repeat_noise)
         # no noise when t == 0
         nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
         return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
 
-    @torch.inference_mode()
+    @torch.no_grad()
-    def p_sample_loop(self, shape, text_cond):
+    def p_sample_loop(self, shape, text_cond, cond_scale = 1.):
         device = self.betas.device
 
         b = shape[0]
@@ -954,7 +960,7 @@ class DiffusionPrior(BaseGaussianDiffusion):
 
         for i in tqdm(reversed(range(0, self.num_timesteps)), desc='sampling loop time step', total=self.num_timesteps):
             times = torch.full((b,), i, device = device, dtype = torch.long)
-            image_embed = self.p_sample(image_embed, times, text_cond = text_cond)
+            image_embed = self.p_sample(image_embed, times, text_cond = text_cond, cond_scale = cond_scale)
 
         return image_embed
 
@@ -978,21 +984,21 @@ class DiffusionPrior(BaseGaussianDiffusion):
         loss = self.loss_fn(pred, target)
         return loss
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @eval_decorator
-    def sample_batch_size(self, batch_size, text_cond):
+    def sample_batch_size(self, batch_size, text_cond, cond_scale = 1.):
         device = self.betas.device
         shape = (batch_size, self.image_embed_dim)
 
         img = torch.randn(shape, device = device)
 
         for i in tqdm(reversed(range(0, self.num_timesteps)), desc = 'sampling loop time step', total = self.num_timesteps):
-            img = self.p_sample(img, torch.full((batch_size,), i, device = device, dtype = torch.long), text_cond = text_cond)
+            img = self.p_sample(img, torch.full((batch_size,), i, device = device, dtype = torch.long), text_cond = text_cond, cond_scale = cond_scale)
         return img
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @eval_decorator
-    def sample(self, text, num_samples_per_batch = 2):
+    def sample(self, text, num_samples_per_batch = 2, cond_scale = 1.):
         # in the paper, what they did was
         # sample 2 image embeddings, choose the top 1 similarity, as judged by CLIP
         text = repeat(text, 'b ... -> (b r) ...', r = num_samples_per_batch)
@@ -1007,7 +1013,7 @@ class DiffusionPrior(BaseGaussianDiffusion):
         if self.condition_on_text_encodings:
             text_cond = {**text_cond, 'text_encodings': text_encodings, 'mask': text_mask}
 
-        image_embeds = self.p_sample_loop((batch_size, image_embed_dim), text_cond = text_cond)
+        image_embeds = self.p_sample_loop((batch_size, image_embed_dim), text_cond = text_cond, cond_scale = cond_scale)
 
         # retrieve original unscaled image embed
 
@@ -1793,6 +1799,7 @@ class Decoder(BaseGaussianDiffusion):
 
         self.image_cond_drop_prob = image_cond_drop_prob
         self.text_cond_drop_prob = text_cond_drop_prob
+        self.can_classifier_guidance = image_cond_drop_prob > 0. or text_cond_drop_prob > 0.
 
         # whether to clip when sampling
 
@@ -1812,13 +1819,19 @@ class Decoder(BaseGaussianDiffusion):
             unet = self.get_unet(unet_number)
 
         self.cuda()
-        self.unets.cpu()
 
+        devices = [module_device(unet) for unet in self.unets]
+        self.unets.cpu()
         unet.cuda()
+
         yield
-        unet.cpu()
+
+        for unet, device in zip(self.unets, devices):
+            unet.to(device)
 
     def p_mean_variance(self, unet, x, t, image_embed, text_encodings = None, text_mask = None, lowres_cond_img = None, clip_denoised = True, predict_x_start = False, learned_variance = False, cond_scale = 1., model_output = None):
+        assert not (cond_scale != 1. and not self.can_classifier_guidance), 'the decoder was not trained with conditional dropout, and thus one cannot use classifier free guidance (cond_scale anything other than 1)'
+
         pred = default(model_output, lambda: unet.forward_with_cond_scale(x, t, image_embed = image_embed, text_encodings = text_encodings, text_mask = text_mask, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img))
 
         if learned_variance:
@@ -1847,7 +1860,7 @@ class Decoder(BaseGaussianDiffusion):
 
         return model_mean, posterior_variance, posterior_log_variance
 
-    @torch.inference_mode()
+    @torch.no_grad()
     def p_sample(self, unet, x, t, image_embed, text_encodings = None, text_mask = None, cond_scale = 1., lowres_cond_img = None, predict_x_start = False, learned_variance = False, clip_denoised = True, repeat_noise = False):
         b, *_, device = *x.shape, x.device
         model_mean, _, model_log_variance = self.p_mean_variance(unet, x = x, t = t, image_embed = image_embed, text_encodings = text_encodings, text_mask = text_mask, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img, clip_denoised = clip_denoised, predict_x_start = predict_x_start, learned_variance = learned_variance)
@@ -1856,14 +1869,15 @@ class Decoder(BaseGaussianDiffusion):
         nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
         return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
 
-    @torch.inference_mode()
+    @torch.no_grad()
-    def p_sample_loop(self, unet, shape, image_embed, predict_x_start = False, learned_variance = False, clip_denoised = True, lowres_cond_img = None, text_encodings = None, text_mask = None, cond_scale = 1):
+    def p_sample_loop(self, unet, shape, image_embed, predict_x_start = False, learned_variance = False, clip_denoised = True, lowres_cond_img = None, text_encodings = None, text_mask = None, cond_scale = 1, is_latent_diffusion = False):
         device = self.betas.device
 
         b = shape[0]
         img = torch.randn(shape, device = device)
 
-        lowres_cond_img = maybe(normalize_neg_one_to_one)(lowres_cond_img)
+        if not is_latent_diffusion:
+            lowres_cond_img = maybe(normalize_neg_one_to_one)(lowres_cond_img)
 
         for i in tqdm(reversed(range(0, self.num_timesteps)), desc = 'sampling loop time step', total = self.num_timesteps):
             img = self.p_sample(
@@ -1883,13 +1897,14 @@ class Decoder(BaseGaussianDiffusion):
         unnormalize_img = unnormalize_zero_to_one(img)
         return unnormalize_img
 
-    def p_losses(self, unet, x_start, times, *, image_embed, lowres_cond_img = None, text_encodings = None, text_mask = None, predict_x_start = False, noise = None, learned_variance = False, clip_denoised = False):
+    def p_losses(self, unet, x_start, times, *, image_embed, lowres_cond_img = None, text_encodings = None, text_mask = None, predict_x_start = False, noise = None, learned_variance = False, clip_denoised = False, is_latent_diffusion = False):
         noise = default(noise, lambda: torch.randn_like(x_start))
 
         # normalize to [-1, 1]
 
-        x_start = normalize_neg_one_to_one(x_start)
+        if not is_latent_diffusion:
-        lowres_cond_img = maybe(normalize_neg_one_to_one)(lowres_cond_img)
+            x_start = normalize_neg_one_to_one(x_start)
+            lowres_cond_img = maybe(normalize_neg_one_to_one)(lowres_cond_img)
 
         # get x_t
 
@@ -1949,12 +1964,14 @@ class Decoder(BaseGaussianDiffusion):
 
         return loss + vb_loss
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @eval_decorator
     def sample(
         self,
         image_embed = None,
         text = None,
+        text_mask = None,
+        text_encodings = None,
         batch_size = 1,
         cond_scale = 1.,
         stop_at_unet_number = None
@@ -1964,8 +1981,8 @@ class Decoder(BaseGaussianDiffusion):
         if not self.unconditional:
             batch_size = image_embed.shape[0]
 
-        text_encodings = text_mask = None
+        if exists(text) and not exists(text_encodings) and not self.unconditional:
-        if exists(text):
+            assert exists(self.clip)
             _, text_encodings, text_mask = self.clip.embed_text(text)
 
         assert not (self.condition_on_text_encodings and not exists(text_encodings)), 'text or text encodings must be passed into decoder if specified'
@@ -2001,7 +2018,8 @@ class Decoder(BaseGaussianDiffusion):
                     predict_x_start = predict_x_start,
                     learned_variance = learned_variance,
                     clip_denoised = not is_latent_diffusion,
-                    lowres_cond_img = lowres_cond_img
+                    lowres_cond_img = lowres_cond_img,
+                    is_latent_diffusion = is_latent_diffusion
                 )
 
                 img = vae.decode(img)
@@ -2017,6 +2035,7 @@ class Decoder(BaseGaussianDiffusion):
         text = None,
         image_embed = None,
         text_encodings = None,
+        text_mask = None,
         unet_number = None
     ):
         assert not (len(self.unets) > 1 and not exists(unet_number)), f'you must specify which unet you want trained, from a range of 1 to {len(self.unets)}, if you are training cascading DDPM (multiple unets)'
@@ -2041,7 +2060,6 @@ class Decoder(BaseGaussianDiffusion):
             assert exists(self.clip), 'if you want to derive CLIP image embeddings automatically, you must supply `clip` to the decoder on init'
             image_embed, _ = self.clip.embed_image(image)
 
-        text_encodings = text_mask = None
         if exists(text) and not exists(text_encodings) and not self.unconditional:
             assert exists(self.clip), 'if you are passing in raw text, you need to supply `clip` to the decoder'
             _, text_encodings, text_mask = self.clip.embed_text(text)
@@ -2060,12 +2078,14 @@ class Decoder(BaseGaussianDiffusion):
             image = aug(image)
             lowres_cond_img = aug(lowres_cond_img, params = aug._params)
 
+        is_latent_diffusion = not isinstance(vae, NullVQGanVAE)
+
         vae.eval()
         with torch.no_grad():
             image = vae.encode(image)
             lowres_cond_img = maybe(vae.encode)(lowres_cond_img)
 
-        return self.p_losses(unet, image, times, image_embed = image_embed, text_encodings = text_encodings, text_mask = text_mask, lowres_cond_img = lowres_cond_img, predict_x_start = predict_x_start, learned_variance = learned_variance)
+        return self.p_losses(unet, image, times, image_embed = image_embed, text_encodings = text_encodings, text_mask = text_mask, lowres_cond_img = lowres_cond_img, predict_x_start = predict_x_start, learned_variance = learned_variance, is_latent_diffusion = is_latent_diffusion)
 
 # main class
 
@@ -2088,22 +2108,23 @@ class DALLE2(nn.Module):
 
         self.to_pil = T.ToPILImage()
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @eval_decorator
     def forward(
         self,
         text,
         cond_scale = 1.,
+        prior_cond_scale = 1.,
         return_pil_images = False
     ):
-        device = next(self.parameters()).device
+        device = module_device(self)
         one_text = isinstance(text, str) or (not is_list_str(text) and text.shape[0] == 1)
 
         if isinstance(text, str) or is_list_str(text):
             text = [text] if not isinstance(text, (list, tuple)) else text
             text = tokenizer.tokenize(text).to(device)
 
-        image_embed = self.prior.sample(text, num_samples_per_batch = self.prior_num_samples)
+        image_embed = self.prior.sample(text, num_samples_per_batch = self.prior_num_samples, cond_scale = prior_cond_scale)
 
         text_cond = text if self.decoder_need_text_cond else None
         images = self.decoder.sample(image_embed, text = text_cond, cond_scale = cond_scale)

dalle2_pytorch/dataloaders/__init__.py

@@ -1 +1,2 @@
-from dalle2_pytorch.dataloaders.decoder_loader import ImageEmbeddingDataset, create_image_embedding_dataloader
+from dalle2_pytorch.dataloaders.decoder_loader import ImageEmbeddingDataset, create_image_embedding_dataloader
+from dalle2_pytorch.dataloaders.embedding_wrapper import make_splits

dalle2_pytorch/dataloaders/embedding_wrapper.py

@@ -0,0 +1,180 @@
+from torch.utils.data import IterableDataset
+from torch import from_numpy
+from clip import tokenize
+from embedding_reader import EmbeddingReader
+
+
+class PriorEmbeddingLoader(IterableDataset):
+    def __init__(
+        self,
+        text_conditioned: bool,
+        batch_size: int,
+        start: int,
+        stop: int,
+        image_reader,
+        text_reader: EmbeddingReader = None,
+        device: str = "cpu",
+    ) -> None:
+        super(PriorEmbeddingLoader).__init__()
+
+        self.text_conditioned = text_conditioned
+
+        if not self.text_conditioned:
+            self.text_reader = text_reader
+
+        self.image_reader = image_reader
+        self.batch_size = batch_size
+        self.start = start
+        self.stop = stop
+        self.device = device
+
+    def __iter__(self):
+        self.n = 0
+        loader_args = dict(
+            batch_size=self.batch_size,
+            start=self.start,
+            end=self.stop,
+            show_progress=False,
+        )
+        if self.text_conditioned:
+            self.loader = self.image_reader(**loader_args)
+        else:
+            self.loader = zip(
+                self.image_reader(**loader_args), self.text_reader(**loader_args)
+            )
+        return self
+
+    def __next__(self):
+        try:
+            return self.get_sample()
+        except StopIteration:
+            raise StopIteration
+
+    def get_sample(self):
+        """
+        pre-proocess data from either reader into a common format
+        """
+        self.n += 1
+
+        if self.text_conditioned:
+            image_embedding, caption = next(self.loader)
+
+            image_embedding = from_numpy(image_embedding).to(self.device)
+            tokenized_caption = tokenize(
+                caption["caption"].to_list(), truncate=True
+            ).to(self.device)
+
+            return image_embedding, tokenized_caption
+
+        else:
+            (image_embedding, _), (text_embedding, _) = next(self.loader)
+
+            image_embedding = from_numpy(image_embedding).to(self.device)
+            text_embedding = from_numpy(text_embedding).to(self.device)
+
+            return image_embedding, text_embedding
+
+
+def make_splits(
+    text_conditioned: bool,
+    batch_size: int,
+    num_data_points: int,
+    train_split: float,
+    eval_split: float,
+    device: str,
+    img_url: str,
+    meta_url: str = None,
+    txt_url: str = None,
+):
+
+    assert img_url is not None, "Must supply some image embeddings"
+
+    if text_conditioned:
+        assert meta_url is not None, "Must supply metadata url if text-conditioning"
+        image_reader = EmbeddingReader(
+            embeddings_folder=img_url,
+            file_format="parquet_npy",
+            meta_columns=["caption"],
+            metadata_folder=meta_url,
+        )
+
+        # compute split points
+        if num_data_points > image_reader.count:
+            print("Specified point count is larger than the number of points available...defaulting to max length of reader.")
+            num_data_points = image_reader.count
+
+        train_set_size = int(train_split * num_data_points)
+        eval_set_size = int(eval_split * num_data_points)
+        eval_stop = int(train_set_size + eval_set_size)
+
+        train_loader = PriorEmbeddingLoader(
+            text_conditioned=text_conditioned,
+            image_reader=image_reader,
+            batch_size=batch_size,
+            start=0,
+            stop=train_set_size,
+            device=device,
+        )
+        eval_loader = PriorEmbeddingLoader(
+            text_conditioned=text_conditioned,
+            image_reader=image_reader,
+            batch_size=batch_size,
+            start=train_set_size,
+            stop=eval_stop,
+            device=device,
+        )
+        test_loader = PriorEmbeddingLoader(
+            text_conditioned=text_conditioned,
+            image_reader=image_reader,
+            batch_size=batch_size,
+            start=eval_stop,
+            stop=int(num_data_points),
+            device=device,
+        )
+
+    else:
+        assert (
+            txt_url is not None
+        ), "Must supply text embedding url if not text-conditioning"
+
+        image_reader = EmbeddingReader(img_url, file_format="npy")
+        text_reader = EmbeddingReader(txt_url, file_format="npy")
+
+        # compute split points
+        if num_data_points > image_reader.count:
+            print("Specified point count is larger than the number of points available...defaulting to max length of reader.")
+            num_data_points = image_reader.count
+
+        train_set_size = int(train_split * num_data_points)
+        eval_set_size = int(eval_split * num_data_points)
+        eval_stop = int(train_set_size + eval_set_size)
+
+        train_loader = PriorEmbeddingLoader(
+            text_conditioned=text_conditioned,
+            image_reader=image_reader,
+            text_reader=text_reader,
+            batch_size=batch_size,
+            start=0,
+            stop=train_set_size,
+            device=device,
+        )
+        eval_loader = PriorEmbeddingLoader(
+            text_conditioned=text_conditioned,
+            image_reader=image_reader,
+            text_reader=text_reader,
+            batch_size=batch_size,
+            start=train_set_size,
+            stop=eval_stop,
+            device=device,
+        )
+        test_loader = PriorEmbeddingLoader(
+            text_conditioned=text_conditioned,
+            image_reader=image_reader,
+            text_reader=text_reader,
+            batch_size=batch_size,
+            start=eval_stop,
+            stop=int(num_data_points),
+            device=device,
+        )
+
+    return train_loader, eval_loader, test_loader

dalle2_pytorch/dataloaders/simple_image_only_dataloader.py

@@ -0,0 +1,59 @@
+from pathlib import Path
+
+import torch
+from torch.utils import data
+from torchvision import transforms, utils
+
+from PIL import Image
+
+# helpers functions
+
+def cycle(dl):
+    while True:
+        for data in dl:
+            yield data
+
+# dataset and dataloader
+
+class Dataset(data.Dataset):
+    def __init__(
+        self,
+        folder,
+        image_size,
+        exts = ['jpg', 'jpeg', 'png']
+    ):
+        super().__init__()
+        self.folder = folder
+        self.image_size = image_size
+        self.paths = [p for ext in exts for p in Path(f'{folder}').glob(f'**/*.{ext}')]
+
+        self.transform = transforms.Compose([
+            transforms.Resize(image_size),
+            transforms.RandomHorizontalFlip(),
+            transforms.CenterCrop(image_size),
+            transforms.ToTensor()
+        ])
+
+    def __len__(self):
+        return len(self.paths)
+
+    def __getitem__(self, index):
+        path = self.paths[index]
+        img = Image.open(path)
+        return self.transform(img)
+
+def get_images_dataloader(
+    folder,
+    *,
+    batch_size,
+    image_size,
+    shuffle = True,
+    cycle_dl = True,
+    pin_memory = True
+):
+    ds = Dataset(folder, image_size)
+    dl = data.DataLoader(ds, batch_size = batch_size, shuffle = shuffle, pin_memory = pin_memory)
+
+    if cycle_dl:
+        dl = cycle(dl)
+    return dl

dalle2_pytorch/trainer.py

@@ -179,8 +179,8 @@ class EMA(nn.Module):
         self.online_model = model
         self.ema_model = copy.deepcopy(model)
 
-        self.update_after_step = update_after_step # only start EMA after this step number, starting at 0
         self.update_every = update_every
+        self.update_after_step = update_after_step  // update_every # only start EMA after this step number, starting at 0
 
         self.register_buffer('initted', torch.Tensor([False]))
         self.register_buffer('step', torch.tensor([0.]))
@@ -189,6 +189,9 @@ class EMA(nn.Module):
         device = self.initted.device
         self.ema_model.to(device)
 
+    def copy_params_from_model_to_ema(self):
+        self.ema_model.state_dict(self.online_model.state_dict())
+
     def update(self):
         self.step += 1
 
@@ -196,7 +199,7 @@ class EMA(nn.Module):
             return
 
         if not self.initted:
-            self.ema_model.state_dict(self.online_model.state_dict())
+            self.copy_params_from_model_to_ema()
             self.initted.data.copy_(torch.Tensor([True]))
 
         self.update_moving_average(self.ema_model, self.online_model)
@@ -278,17 +281,17 @@ class DiffusionPriorTrainer(nn.Module):
 
         self.step += 1
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @cast_torch_tensor
     def p_sample_loop(self, *args, **kwargs):
         return self.ema_diffusion_prior.ema_model.p_sample_loop(*args, **kwargs)
 
-    @torch.inference_mode()
+    @torch.no_grad()
     @cast_torch_tensor
     def sample(self, *args, **kwargs):
         return self.ema_diffusion_prior.ema_model.sample(*args, **kwargs)
 
-    @torch.inference_mode()
+    @torch.no_grad()
     def sample_batch_size(self, *args, **kwargs):
         return self.ema_diffusion_prior.ema_model.sample_batch_size(*args, **kwargs)
 
@@ -377,8 +380,11 @@ class DecoderTrainer(nn.Module):
         scaler = getattr(self, f'scaler{index}')
         return scaler.scale(loss)
 
-    def update(self, unet_number):
+    def update(self, unet_number = None):
-        assert 1 <= unet_number <= self.num_unets
+        if self.num_unets == 1:
+            unet_number = default(unet_number, 1)
+
+        assert exists(unet_number) and 1 <= unet_number <= self.num_unets
         index = unet_number - 1
         unet = self.decoder.unets[index]
 
@@ -402,6 +408,9 @@ class DecoderTrainer(nn.Module):
     @torch.no_grad()
     @cast_torch_tensor
     def sample(self, *args, **kwargs):
+        if kwargs.pop('use_non_ema', False):
+            return self.decoder.sample(*args, **kwargs)
+
         if self.use_ema:
             trainable_unets = self.decoder.unets
             self.decoder.unets = self.unets                  # swap in exponential moving averaged unets for sampling
@@ -421,10 +430,13 @@ class DecoderTrainer(nn.Module):
     def forward(
         self,
         *args,
-        unet_number,
+        unet_number = None,
         max_batch_size = None,
         **kwargs
     ):
+        if self.num_unets == 1:
+            unet_number = default(unet_number, 1)
+
         total_loss = 0.
 
         for chunk_size_frac, (chunked_args, chunked_kwargs) in split_args_and_kwargs(*args, split_size = max_batch_size, **kwargs):

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.2.37',
+  version = '0.2.43',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',

train_diffusion_prior.py

@@ -5,10 +5,13 @@ import time
 import numpy as np
 
 import torch
+import clip
 from torch import nn
 
-from dalle2_pytorch import DiffusionPrior, DiffusionPriorNetwork
+from dalle2_pytorch.dataloaders import make_splits
-from dalle2_pytorch.train import DiffusionPriorTrainer, load_diffusion_model, save_diffusion_model, print_ribbon
+from dalle2_pytorch import DiffusionPrior, DiffusionPriorNetwork, OpenAIClipAdapter
+from dalle2_pytorch.trainer import DiffusionPriorTrainer, load_diffusion_model, save_diffusion_model, print_ribbon
+
 from dalle2_pytorch.trackers import ConsoleTracker, WandbTracker
 
 from embedding_reader import EmbeddingReader
@@ -17,8 +20,7 @@ from tqdm import tqdm
 
 # constants
 
-NUM_TEST_EMBEDDINGS = 100 # for cosine similarity reporting during training
+REPORT_METRICS_EVERY = 250 # for cosine similarity and other metric reporting during training
-REPORT_METRICS_EVERY = 100 # for cosine similarity and other metric reporting during training
 
 tracker = WandbTracker()
 
@@ -36,81 +38,106 @@ class Timer:
 
     def elapsed(self):
         return time.time() - self.last_time
+
 # functions
 
-def eval_model(model,device,image_reader,text_reader,start,end,batch_size,loss_type,phase="Validation"):
+def eval_model(model, dataloader, text_conditioned, loss_type, phase="Validation"):
     model.eval()
+
     with torch.no_grad():
         total_loss = 0.
         total_samples = 0.
 
-        for emb_images, emb_text in zip(image_reader(batch_size=batch_size, start=start, end=end),
+        for image_embeddings, text_data in tqdm(dataloader):
-                text_reader(batch_size=batch_size, start=start, end=end)):
 
-            emb_images_tensor = torch.tensor(emb_images[0]).to(device)
+            batches = image_embeddings.shape[0]
-            emb_text_tensor = torch.tensor(emb_text[0]).to(device)
 
-            batches = emb_images_tensor.shape[0]
+            input_args = dict(image_embed=image_embeddings)
+            if text_conditioned:
+                input_args = dict(**input_args, text = text_data)
+            else:
+                input_args = dict(**input_args, text_embed=text_data)
 
-            loss = model(text_embed = emb_text_tensor, image_embed = emb_images_tensor)
+            loss = model(**input_args)
 
-            total_loss += loss.item() * batches
+            total_loss += loss * batches
             total_samples += batches
 
         avg_loss = (total_loss / total_samples)
+
         tracker.log({f'{phase} {loss_type}': avg_loss})
 
-def report_cosine_sims(diffusion_prior,image_reader,text_reader,train_set_size,NUM_TEST_EMBEDDINGS,device):
+def report_cosine_sims(diffusion_prior, dataloader, text_conditioned):
     diffusion_prior.eval()
 
     cos = nn.CosineSimilarity(dim=1, eps=1e-6)
 
-    tstart = train_set_size
+    for test_image_embeddings, text_data in tqdm(dataloader):
-    tend = train_set_size+NUM_TEST_EMBEDDINGS
+
+        # we are text conditioned, we produce an embedding from the tokenized text
+        if text_conditioned:
+            text_embedding, text_encodings, text_mask = diffusion_prior.clip.embed_text(
+                text_data)
+            text_cond = dict(text_embed=text_embedding,
+                             text_encodings=text_encodings, mask=text_mask)
+        else:
+            text_embedding = text_data
+            text_cond = dict(text_embed=text_embedding)
+
+        # make a copy of the text embeddings for shuffling
+        text_embed_shuffled = text_embedding.clone()
+
+        # roll the text to simulate "unrelated" captions
+        rolled_idx = torch.roll(torch.arange(text_embedding.shape[0]), 1)
+        text_embed_shuffled = text_embed_shuffled[rolled_idx]
+        text_embed_shuffled = text_embed_shuffled / \
+            text_embed_shuffled.norm(dim=1, keepdim=True)
+
+        if text_conditioned:
+            text_encodings_shuffled = text_encodings[rolled_idx]
+            text_mask_shuffled = text_mask[rolled_idx]
+        else:
+            text_encodings_shuffled = None
+            text_mask_shuffled = None
+
+        text_cond_shuffled = dict(text_embed=text_embed_shuffled,
+                                  text_encodings=text_encodings_shuffled, mask=text_mask_shuffled)
 
-    for embt, embi in zip(text_reader(batch_size=NUM_TEST_EMBEDDINGS, start=tstart, end=tend), 
-            image_reader(batch_size=NUM_TEST_EMBEDDINGS, start=tstart, end=tend)):
-       # make a copy of the text embeddings for shuffling
-       text_embed = torch.tensor(embt[0]).to(device)
-       text_embed_shuffled = text_embed.clone()
-        # roll the text embeddings to simulate "unrelated" captions
-       rolled_idx = torch.roll(torch.arange(NUM_TEST_EMBEDDINGS), 1)
-       text_embed_shuffled = text_embed_shuffled[rolled_idx]
-       text_embed_shuffled = text_embed_shuffled / \
-           text_embed_shuffled.norm(dim=1, keepdim=True)
-       test_text_shuffled_cond = dict(text_embed=text_embed_shuffled)
         # prepare the text embedding
-       text_embed = text_embed / text_embed.norm(dim=1, keepdim=True)
+        text_embed = text_embedding / text_embedding.norm(dim=1, keepdim=True)
-       test_text_cond = dict(text_embed=text_embed)
+
         # prepare image embeddings
-       test_image_embeddings = torch.tensor(embi[0]).to(device)
+        test_image_embeddings = test_image_embeddings / \
-       test_image_embeddings = test_image_embeddings / \
+            test_image_embeddings.norm(dim=1, keepdim=True)
-           test_image_embeddings.norm(dim=1, keepdim=True)
+
         # predict on the unshuffled text embeddings
-       predicted_image_embeddings = diffusion_prior.p_sample_loop(
+        predicted_image_embeddings = diffusion_prior.p_sample_loop(
-           (NUM_TEST_EMBEDDINGS, 768), text_cond=test_text_cond)
+            test_image_embeddings.shape, text_cond)
-       predicted_image_embeddings = predicted_image_embeddings / \
+        predicted_image_embeddings = predicted_image_embeddings / \
-           predicted_image_embeddings.norm(dim=1, keepdim=True)
+            predicted_image_embeddings.norm(dim=1, keepdim=True)
+
         # predict on the shuffled embeddings
-       predicted_unrelated_embeddings = diffusion_prior.p_sample_loop(
+        predicted_unrelated_embeddings = diffusion_prior.p_sample_loop(
-           (NUM_TEST_EMBEDDINGS, 768), text_cond=test_text_shuffled_cond)
+            test_image_embeddings.shape, text_cond_shuffled)
-       predicted_unrelated_embeddings = predicted_unrelated_embeddings / \
+        predicted_unrelated_embeddings = predicted_unrelated_embeddings / \
-           predicted_unrelated_embeddings.norm(dim=1, keepdim=True)
+            predicted_unrelated_embeddings.norm(dim=1, keepdim=True)
+
         # calculate similarities
-       original_similarity = cos(
+        original_similarity = cos(
            text_embed, test_image_embeddings).cpu().numpy()
-       predicted_similarity = cos(
+        predicted_similarity = cos(
            text_embed, predicted_image_embeddings).cpu().numpy()
-       unrelated_similarity = cos(
+        unrelated_similarity = cos(
            text_embed, predicted_unrelated_embeddings).cpu().numpy()
-       predicted_img_similarity = cos(
+        predicted_img_similarity = cos(
            test_image_embeddings, predicted_image_embeddings).cpu().numpy()
-       tracker.log({"CosineSimilarity(text_embed,image_embed)": np.mean(original_similarity),
+        tracker.log({"CosineSimilarity(text_embed,image_embed)": np.mean(original_similarity),
             "CosineSimilarity(text_embed,predicted_image_embed)":np.mean(predicted_similarity),
             "CosineSimilarity(orig_image_embed,predicted_image_embed)":np.mean(predicted_img_similarity),
             "CosineSimilarity(text_embed,predicted_unrelated_embed)": np.mean(unrelated_similarity),
             "Cosine similarity difference":np.mean(predicted_similarity - original_similarity)})
 
+
 @click.command()
 @click.option("--wandb-entity", default="laion")
 @click.option("--wandb-project", default="diffusion-prior")
@@ -118,29 +145,32 @@ def report_cosine_sims(diffusion_prior,image_reader,text_reader,train_set_size,N
 @click.option("--wandb-arch", default="DiffusionPrior")
 @click.option("--image-embed-url", default="https://mystic.the-eye.eu/public/AI/cah/laion5b/embeddings/laion2B-en/img_emb/")
 @click.option("--text-embed-url", default="https://mystic.the-eye.eu/public/AI/cah/laion5b/embeddings/laion2B-en/text_emb/")
+@click.option("--meta-url", default="https://mystic.the-eye.eu/public/AI/cah/laion5b/embeddings/laion2B-en/laion2B-en-metadata/")
 @click.option("--learning-rate", default=1.1e-4)
 @click.option("--weight-decay", default=6.02e-2)
 @click.option("--dropout", default=5e-2)
 @click.option("--max-grad-norm", default=0.5)
-@click.option("--batch-size", default=10**4)
+@click.option("--num-data-points", default=250e6)
+@click.option("--batch-size", default=320)
 @click.option("--num-epochs", default=5)
 @click.option("--image-embed-dim", default=768)
-@click.option("--train-percent", default=0.7)
+@click.option("--train-percent", default=0.9)
-@click.option("--val-percent", default=0.2)
+@click.option("--val-percent", default=1e-7)
-@click.option("--test-percent", default=0.1)
+@click.option("--test-percent", default=0.0999999)
-@click.option("--dpn-depth", default=6)
+@click.option("--dpn-depth", default=12)
 @click.option("--dpn-dim-head", default=64)
-@click.option("--dpn-heads", default=8)
+@click.option("--dpn-heads", default=12)
-@click.option("--dp-condition-on-text-encodings", default=False)
+@click.option("--dp-condition-on-text-encodings", default=True)
-@click.option("--dp-timesteps", default=100)
+@click.option("--dp-timesteps", default=1000)
-@click.option("--dp-normformer", default=False)
+@click.option("--dp-normformer", default=True)
 @click.option("--dp-cond-drop-prob", default=0.1)
 @click.option("--dp-loss-type", default="l2")
-@click.option("--clip", default=None)
+@click.option("--clip", default="ViT-L/14")
 @click.option("--amp", default=False)
-@click.option("--save-interval", default=30)
+@click.option("--save-interval", default=120)
 @click.option("--save-path", default="./diffusion_prior_checkpoints")
 @click.option("--pretrained-model-path", default=None)
+@click.option("--gpu-device", default=0)
 def train(
     wandb_entity,
     wandb_project,
@@ -148,10 +178,12 @@ def train(
     wandb_arch,
     image_embed_url,
     text_embed_url,
+    meta_url,
     learning_rate,
     weight_decay,
     dropout,
     max_grad_norm,
+    num_data_points,
     batch_size,
     num_epochs,
     image_embed_dim,
@@ -170,7 +202,8 @@ def train(
     amp,
     save_interval,
     save_path,
-    pretrained_model_path
+    pretrained_model_path,
+    gpu_device
 ):
     config = {
         "learning_rate": learning_rate,
@@ -197,7 +230,7 @@ def train(
 
     # Check if DPRIOR_PATH exists(saved model path)
 
-    DPRIOR_PATH = args.pretrained_model_path
+    DPRIOR_PATH = pretrained_model_path
     RESUME = exists(DPRIOR_PATH)
 
     if not RESUME:
@@ -211,7 +244,7 @@ def train(
 
     has_cuda = torch.cuda.is_available()
     if has_cuda:
-        device = torch.device("cuda:0")
+        device = torch.device(f"cuda:{gpu_device}")
         torch.cuda.set_device(device)
 
     # Training loop
@@ -227,11 +260,17 @@ def train(
         normformer = dp_normformer
     )
     
+    # Load clip model if text-conditioning
+    if dp_condition_on_text_encodings:
+        clip_adapter = OpenAIClipAdapter(clip)
+    else:
+        clip_adapter = None
+        
     # diffusion prior with text embeddings and image embeddings pre-computed
 
     diffusion_prior = DiffusionPrior( 
         net = prior_network,
-        clip = clip,
+        clip = clip_adapter,
         image_embed_dim = image_embed_dim,
         timesteps = dp_timesteps,
         cond_drop_prob = dp_cond_drop_prob,
@@ -265,33 +304,37 @@ def train(
 
     Path(save_path).mkdir(exist_ok = True, parents = True)
 
-    # Get image and text embeddings from the servers
+    # Utilize wrapper to abstract away loader logic
+    print_ribbon("Downloading Embeddings")
+    loader_args = dict(text_conditioned=dp_condition_on_text_encodings, batch_size=batch_size, num_data_points=num_data_points,
+                       train_split=train_percent, eval_split=val_percent, device=device, img_url=image_embed_url)
 
-    print_ribbon("Downloading embeddings - image and text")
+    if dp_condition_on_text_encodings:
-    image_reader = EmbeddingReader(embeddings_folder=image_embed_url, file_format="npy")
+        loader_args = dict(**loader_args, meta_url=meta_url)
-    text_reader  = EmbeddingReader(embeddings_folder=text_embed_url, file_format="npy")
+    else:
-    num_data_points = text_reader.count
+        loader_args = dict(**loader_args, txt_url=text_embed_url)
+
+    train_loader, eval_loader, test_loader = make_splits(**loader_args)
 
     ### Training code ###
 
+    step = 1 
     timer = Timer()
     epochs = num_epochs
 
-    train_set_size = int(train_percent*num_data_points)
-    val_set_size = int(val_percent*num_data_points)
-    eval_start = train_set_size
-
     for _ in range(epochs):
 
-        for emb_images,emb_text in zip(image_reader(batch_size=batch_size, start=0, end=train_set_size),
+        for image, text in tqdm(train_loader):
-                text_reader(batch_size=batch_size, start=0, end=train_set_size)):
-
-            trainer.train()
             
-            emb_images_tensor = torch.tensor(emb_images[0]).to(device)
+            diffusion_prior.train()
-            emb_text_tensor = torch.tensor(emb_text[0]).to(device)
+            
+            input_args = dict(image_embed=image)
+            if dp_condition_on_text_encodings:
+                input_args = dict(**input_args, text = text)
+            else:
+                input_args = dict(**input_args, text_embed=text)
 
-            loss = trainer(text_embed = emb_text_tensor, image_embed = emb_images_tensor)
+            loss = trainer(**input_args)
 
             # Samples per second
 
@@ -310,37 +353,23 @@ def train(
                     image_embed_dim)
 
             # Log to wandb
-            tracker.log({"Training loss": loss.item(),
+            tracker.log({"Training loss": loss,
                         "Steps": step,
                         "Samples per second": samples_per_sec})
             # Log cosineSim(text_embed,predicted_image_embed) - cosineSim(text_embed,image_embed)
             # Use NUM_TEST_EMBEDDINGS samples from the test set each time
             # Get embeddings from the most recently saved model
             if(step % REPORT_METRICS_EVERY) == 0:
-                report_cosine_sims(diffusion_prior,
+                report_cosine_sims(diffusion_prior, eval_loader, dp_condition_on_text_encodings)
-                        image_reader,
-                        text_reader,
-                        train_set_size,
-                        NUM_TEST_EMBEDDINGS,
-                        device)
                 ### Evaluate model(validation run) ###
-                eval_model(diffusion_prior,
+                eval_model(diffusion_prior, eval_loader, dp_condition_on_text_encodings, dp_loss_type, phase="Validation")
-                        device,
-                        image_reader,
-                        text_reader,
-                        eval_start,
-                        eval_start+NUM_TEST_EMBEDDINGS,
-                        NUM_TEST_EMBEDDINGS,
-                        dp_loss_type,
-                        phase="Validation")
 
+            step += 1
             trainer.update()
 
     ### Test run ###
-    test_set_size = int(test_percent*train_set_size) 
+    eval_model(diffusion_prior, test_loader, dp_condition_on_text_encodings, dp_loss_type, phase="Test")
-    start = train_set_size+val_set_size
+
-    end = num_data_points
-    eval_model(diffusion_prior,device,image_reader,text_reader,start,end,batch_size,dp_loss_type,phase="Test")
 
 if __name__ == "__main__":
     train()